Remote Sensing Lab 5: Image Mosaic and Miscellaneous Image Functions 2

Goal
The goal of this lab was to get a beginners grasp at several remote sensing tools. This lab introduced us to image mosaic, spatial and spectral image enhancement, band ratio, and binary change detection.

Methods
The first part of the lab introduced us to image mosaicking. In this section we will take two images of the surrounding Eau Claire area and bring them together. To start we had to overlap the images in the right way and using the mosaic tool (mosaic express) under the raster tab we could start the process of joining the two images together. After joining the two images, since this is not an advanced class we did not change anything away from the defaults, you would get a seamless image of the two satellite obtained images.


Left: Before joining images
Right: After joining the images

After joining the two images without blending the colors you can easily tell where the boundaries are. The next section would work on fixing this. Still working with the mosaic tool, we switched from mosaic express to use MosaicPro instead. Using the same two images, color correction tools, and a histogram matching tool. We were able to join the two images together once again. Again not changing the defaults we got a pretty simple final product.



Left: Before joining the images
Right: After joining the images

After using MosaicPro the color change looked much more natural, apart from the black line down the middle the images look like one.

The next section had to deal with Band Ratios. In this section we will use the NDVI tool under the Raster tab and Unsupervised tool. This tool helps to show where vegetation on the land image is. After inserting the Eau Claire area image and saving it to our own personal files we ran the tool. The end product was a black and white looking image.



Left: Original Image
Right: Image after running NDVI tool to show land use

On the right image the darker areas, not black which represents the river, you can expect to find more urbanized, built up, areas and in the lighter areas find more farmland.

The next section dealt with Spatial and spectral Image enhancement. In this section we deal with high frequency images, images with sharp borders between colors, and low frequency images, images with a more "blurred" looking border between colors. The first part of this section was to apply a 5x5 Low Pass Convolution filter to the image of the Chicago area.



Left: Original Image
Right: Image with a 5x5 Low Pass Convolution Filter

The 5x5 Low Pass Convolution Filter makes the new image to appear smoother than the original. The next section of this part had to deal with applying a 5x5 High Pass Convolution filter to an image in Sierra Leone. Done the same way as the Chicago image, Raster tab> Spatial tool> select Convolution.



Left: Original Image
Right: Image with a 5x5 High Pass Convolution Filter

The new image of Sierra Leone now has much sharper boarders between colors and it also appears much more sharp. The next part of this section was to use a different image of Sierra Leone and apply a 3x3 Laplacian Edge Detection Filter. This filter is used to detect rapid change in an image. From the visual perspective it would look like a quick change of color on the image.

Left: Original Image
Right: New Laplacian Edge Detection

Left: Original zoomed in
Right: New Laplacian Edge Detection filter zoomed in

When zoomed in you can see how different the Laplacian Edge Detection image is from the original. The next section of part 3 dealt with Spectral Enhancement. In this section we will stretch out the color histograms to help make the images look as if they have a wider variety of color to them. For this section we used the Panchromatic tab and the General Contrast tool. After playing around with the variables to get the contrast we wanted we could create a final product that is much easier to interoperate.

The tool used and what the image looked like before applying the tool

After adjusting the contrast and applying it to the same image above

 The last section of part 3 had to deal with Histogram Equalization. Similar to the previous images we are expanding the range of the histograms to add more color to the images. Under the Raster tab> Radiometric tool > Histogram Equalization tool. We then ran the tool to get a much brighter, contrast picture.

Left: Original
Right: New Histogram adjusted image

Even someone who has never seen an image like this before and has no clue what processes have been done to it can tell that the two images look drastically different, some people might even think that they are not even of the same place. In the newly adjusted image compared to the original's histogram you can see a wider area of color being used.

The final part of the lab works with binary change detection also called image differencing. The first part of this lab was to create a difference image. This was done by bringing in two images, one from 1991 and the other from 2011. We then used the functions, two image functions tool, found under the raster tab. After imputing the two images and changing the operation to a - instead of a + and only selecting layer 4, we could save the image to our folder.

Left: Original 1991 Image
Right: Pixels that have changed between 1991 and 2011

After working with the image we then took a look at the metadata and viewed the histogram. Given the mean and standard deviation, we could figure out what portion of the image was in the upper and lower limits. The second section of this part was to map the changes in pixels in the difference image using spatial modeler. This was done using the equation

ΔBV_{ijk = BVijk(1) – BVijk(2) + c}

 







Where
        ΔBV_ijk    = Change in pixel values
        _BVijk(1)    = Brightness value of 2011 image    
         _{BVijk(2)     =  Brightness value of 1991 image}
        C           = constant

In order to find the difference we first had to use model maker. Using just the basic functions we were able to create two different models. The first one dealt with the 2011 Near Infrared band and the 1991 Near Infrared band. We subtracted the 1991 image from the 2011 image and added the constant. The final image would than be used on the next model. The second model was to detect the change/no change threshold value. This model would also use the conditional either if or otherwise function. After running the model we than got an image that showed where the change was. We would later use this image on ArcMap to overlay it on the 1991 Near Infrared band image to see where the changes have occurred.

Results
The results from the last section showed that the pixel changes were in relation to changing lands. Over the past twenty years we have seen changes in areas of urbanization, road creation, farm land change, possible water level changes, and many more features.

Sources
Erdas Imagine 2013
ArcMap 10.2
Images provided by Dr. Wilson